Shadow AI: When Your Team Uses AI in Secret

Shadow AI is when people in your business use AI tools for work that the business never provided or approved, usually a personal ChatGPT, Claude or Gemini account in the browser. The real risk is rarely the AI itself. It is that customer data, quotes and internal details end up in accounts with no contract and no rules behind them. The fix is not a ban but an approved tool, a short rule and an honest conversation with your team.

I am a developer, not a lawyer, and this is not legal advice, just the operator's perspective, under EU and German law. My AI employees also process real customer receipts and customer emails, so for customer data with personal information I use a company plan, not an individual consumer account. The figures in this article come from two German surveys that I read at the source on 28 October 2026. More on AI in day-to-day business is collected on my AI Basics page.

What shadow AI is, and what it isn't

The German Economic Institute (IW) defines shadow AI as the use of AI applications at work that the employer has not provided, such as personal chatbot accounts. It adds a caveat that matters: not provided does not necessarily mean secret or forbidden. It usually turns secret simply because nobody talks about it.

Typical examples from office life:

• An employee rewrites a customer email in her personal chatbot account.
• Someone summarises a contract with a browser extension he installed himself.
• A team member records a client call with a transcription app on their own phone.
• In software you already use, someone switches on a new AI feature without anyone checking what happens to the data.

A tool you have approved is not shadow AI, even if someone uses it clumsily. That is a training question, which I cover in AI Literacy Under the AI Act: Training Staff.

How common shadow AI is

Two German surveys give a picture:

• IW short report no. 63/2026, 19 September 2026. Almost 5,000 employees were surveyed in March and April 2026. Of those who use AI at work, 29.1 percent use tools their employer has not provided. In companies with fewer than 50 employees the figure is 33.5 percent, in companies with 250 or more it is 25.5 percent.
• Bitkom press release, 21 October 2025. Bitkom, Germany's digital industry association, surveyed 604 companies with at least 20 employees. In 8 percent, use of personal AI tools is widespread, 17 percent report isolated cases, and another 17 percent assume it happens without knowing for sure. Only 29 percent are certain it does not happen. 23 percent had set rules for using AI tools.

One finding is especially revealing: according to the IW, shadow AI is most common where employers actively build AI skills. The authors call it a provision gap. People have learned what AI can do and want to use it, but the official tool is missing. In small businesses in particular, shadow AI is less a sign of disobedience than of a missing offer.

Why teams turn to shadow AI

Three causes follow from the two sources. The fourth is my own assessment:

1. There is no approved tool. According to Bitkom, only 23 percent of companies with 20 to 99 employees officially provided generative AI in 2025.
2. There is no rule. If nobody knows what is allowed, everyone decides for themselves.
3. The habit comes from home. The IW suggests that people who regularly use generative AI privately are more likely to carry the habit into work.
4. Nobody is responsible. If it is unclear whom to ask, people don't ask. This is my assessment, not a survey result. How to assign that responsibility without creating a new job is covered in Do I Need an AI Officer?

Shadow AI risks: what can actually go wrong

Data protection. As soon as personal data goes into an AI tool, the GDPR applies. If a provider processes data on your behalf, Article 28 GDPR requires a data processing agreement. Your employee's personal account has no such agreement with your business. Then there is training: on Claude Free, Pro and Max, a setting decides whether chats are used for training, while on Team, Enterprise and the API, data is excluded by default according to Anthropic. Which data belongs in an AI tool at all is covered in AI and Privacy: What the AI Gets to See.

Trade secrets. Under section 2 no. 1(b) of the German Trade Secrets Act, information is only protected if, among other things, it is the subject of reasonable secrecy measures. If price calculations and customer lists drift into personal accounts without any rule, it may be harder to show later that such measures were in place.

Quality. AI output can sound plausible and still be wrong. With shadow AI, nobody checks systematically before something goes to a customer. Why models make things up is explained in AI Hallucinations: Why AI Makes Things Up.

Knowledge that leaves with people. Good prompts, templates and chat histories sit in personal accounts. When the person leaves, so does the knowledge.

AI literacy. Under Article 4 of the EU AI Act, deployers take measures to support the AI literacy of their staff. You can only train people on tools you know about.

How to spot shadow AI without monitoring your team

Start with the mildest option. The first three steps cost nothing and need no technology:

1. Ask, with a promise. A short meeting or a note: which AI tools do you use for work, for what, and what data goes in? Add an explicit promise that honest answers have no consequences. Without that promise, you won't learn much.
2. Check subscriptions and expenses. Expense receipts, company credit cards, app store invoices: AI subscriptions showing up there are a hint, not an accusation.
3. Review the software you already have. Many programs you already use are adding AI features. Check the settings and admin panel to see what is switched on.
4. Don't rely on AI detectors. Whether a text was written with AI cannot be determined reliably. The reasons are in Can You Detect AI Text? What Detectors Do.
5. Technical controls only with care. Network logs or central management of browser extensions show which services are being used. But that means processing your employees' data, and in Germany, if there is a works council, under Federal Labour Court case law such a system requires its consent under section 87(1) no. 6 of the Works Constitution Act if it is objectively capable of monitoring behaviour or performance. What that means in detail is in AI and the Works Council in Germany.

Bringing it under control

Bitkom's president Ralf Wintergerst put it in two parts: businesses need to set clear rules for AI use and provide AI technology to their employees. Both belong together. The IW warns explicitly that outright bans would mainly hit the most engaged employees.

Here is how to go about it:

1. Provide an approved tool. A company account on a plan with a data processing agreement, set up by you, not by each person individually. Where data should not leave the business at all, a local model is an option, which I weigh up in Local AI and Privacy: Cloud or Server? If you already work with Microsoft 365, read Microsoft Copilot and GDPR at Work first.
2. One page of rules. Which tools, which data, who checks. What exactly belongs in it, with a template, is in AI Usage Policy for Small Teams. I deliberately don't repeat the rules here.
3. Move, don't just delete. Ask your team to move good prompts and templates from personal accounts into the company account and to remove company data from personal chat histories. That way the knowledge stays in the business.
4. Train and keep a record. A short briefing per tool, with date and attendees. That is also your documentation for Article 4 of the AI Act.
5. Someone to ask. People who are unsure need a contact, otherwise the next round of shadow AI starts.

Frequently asked questions

Is shadow AI illegal?

There is no general ban. Whether the use is allowed depends on your instructions as the employer and on which data goes in. If personal data goes into a personal account without a legal basis, that can be a data protection breach. This is not legal advice; if in doubt, have your case reviewed professionally.

How do I find out whether my team uses shadow AI?

Ask. An open conversation with a promise that honest answers have no consequences gets you further than any technology. After that, look at subscriptions, expenses and the AI features in the software you already use.

Should I simply ban personal AI accounts?

You can decide that only approved tools may be used for company data, and you should. But a ban without an approved alternative doesn't solve the problem, because the need is still there. According to the IW, outright bans mainly hit the most engaged employees.

What is the difference between shadow AI and shadow IT?

Shadow IT means any software used without approval, and shadow AI is part of it. AI adds two things: the provider processes the content to generate an answer, depending on the plan also for training, and the results flow back into your work unchecked.

Is a company account enough to make shadow AI disappear?

No. A company account removes the most common reason. You also need a rule on which data may go in, a briefing and a person who answers questions.

Where to go from here

Ask your team one question this week: which AI tools do you use for work, and for what? The answers give you the list of tools you want to approve, and then you write the AI usage policy for your team. If you are just bringing AI into your business, Bringing AI to a Small Business helps with the order of steps. How other businesses set up their tools and rules is what we discuss in my community Claude Practitioners. Membership does not replace a legal review.